Home

Algenta SDK Wiki

Python and TypeScript client libraries for Algenta — self-hosted building blocks for AI applications. Source: thyn-ai/algenta-sdk (Apache-2.0). Hosted product docs: docs.algenta.ai.

Algenta runs every workload under an explicit execution policy, and every consequential action is attributable: governed data queries, Monte Carlo simulations, decision memory with execution receipts, agent runs with human-in-the-loop approvals, managed connectors, and a full audit trail — enforced by the engine, never by client-side convention.

Pick your path

I want to…	Start here
Use the SDK in my Python/TS app	Getting Started → Error Handling
Run the examples (ChatGPT, LangGraph, n8n, Mojo…)	Examples → Mojo Quickstart
Understand the platform (contract, device binding, trust boundary)	Architecture → Governance Profiles & Receipts
Contribute a fix, example, or doc	Contributing → Security
Benchmarks & releases (SemVer, signed releases, verify an artifact)	Versioning & Releases

Stuck? Check the FAQ, then GitHub Discussions.

60-second quickstart

pip install algenta-sdk          # Python 3.10+, import name is decision_engine
npm install algenta-sdk          # or TypeScript / Node.js 18+
export ALGENTA_API_KEY=de_live_...   # hosted API key; defaults to https://api.algenta.ai

from decision_engine import AlgentaClient

client = AlgentaClient()  # reads ALGENTA_API_KEY

datasets = client.list_datasets(search="orders", compact=True)
summary = client.get_dataset_summary(datasets.datasets[0].dataset_id)
result = client.query_with_metadata(
    {
        "dataset_id": summary.dataset_id,
        "metric": {"hint": "gross_revenue"},
        "aggregation": "sum",
    }
)
print(result.data.result)

import { AlgentaClient } from "algenta-sdk";

const client = new AlgentaClient(); // reads ALGENTA_API_KEY

const datasets = await client.listDatasets({ search: "orders", compact: true });
const summary = await client.getDatasetSummary(datasets.datasets[0].dataset_id);
const result = await client.queryWithMetadata({
  dataset_id: summary.dataset_id,
  metric: { hint: "gross_revenue" },
  aggregation: "sum",
});
console.log(result.data.result);

Self-hosted engine? Pass base_url="http://localhost:8000" (Python) / baseUrl: "http://localhost:8000" (TS) and your operator-provisioned key. Details: Getting Started.

Architecture in one paragraph

Both SDKs are plain, generated-contract HTTP clients: they discover the engine's published contract at runtime (GET /v1/meta/contract, with an /openapi.json fallback for older self-hosted nodes), and every call is validated against typed request/response models. The TypeScript package additionally ships a Runtime facade for local and self-hosted execution; a separate algenta PyPI package provides the pure-local Python runtime. The engine's compute kernels are proprietary Mojo, distributed as signed algenta-runtime-native wheels — never in this repo. Governance (policies, receipts, approvals, audit) is enforced server-side; the SDKs hold no license keys and no entitlement logic. Full picture: Architecture.

The product core

Four governed tool profiles (observe → govern → execute → full) decide what a model can even see; a successful execute_decision returns a typed execution receipt pinning the policy and schema snapshots it ran under; denials come back synchronously, naming the gate. Read Governance Profiles & Receipts.

Ecosystem

• thyn-ai/algenta-sdk — this repo: Python + TypeScript SDKs, the algenta-mcp MCP server (packages/mcp/algenta_mcp), examples, release metadata
• thyn-ai/algenta-integrations — framework integrations (LangChain, LlamaIndex, pydantic-ai, n8n and more) built on these SDKs
• thyn-ai/mojo-kernels — standalone clean-room Mojo kernels (BM25, fuzzy search, electron-density grids) with drop-in Python/TS packages